1. Field of the Invention
The present invention relates to a color signal processing apparatus suitable for use with a single-plate color camera.
2. Description of Related Background Art
As a signal processing apparatus for a single-plate color camera, a switch-Y system using a color sensor of a pure color stripe type or a complementary color stripe type, a mosaic system using a stripe type color sensor, and the like have been proposed heretofore. The switch-Y system using a stripe color sensor in particular is simple in its circuit arrangement so that it has been widely used. The process conducted in the switch-Y system of a pure color stripe type is shown in FIG. 2 in the form of a block diagram.
Red (R), green (G) and blue (B) signals obtained from R, G and B colors are made to have the same output level relative to a white color at amplifiers 201 to 203. Thereafter, they are subjected to A/D conversion within a gamma correction circuit 204 and then gamma-corrected.
Thereafter, as disclosed in U.S. Pat. No. 4,751,567, the R, G and B signals are alternately selected to form a broad bandwidth luminance signal YH at a switch-Y 208. The bandwidth of signal YH is limited at a low-pass filter 210 and passed to a delay 211 and aperture compensation (APC) filter or contour emphasizing circuit 212 to obtain a final signal Y. In the meantime, the color signals are made narrower at low-pass filters 205 to 207 and inputted to a processor 209 whereat a narrow band luminance signal YL is obtained from the inputted R, G and B signals to output two color difference signals R-YL and B-YL.
Consider next the above-described technique applied to a sensor with a color filter mounted on the pixels disposed in a so-called offset sampling array, as shown in FIG. 3.
In a typical sensor, 640 pixels are disposed on one line in the horizontal direction, and 480 lines are disposed in the vertical direction, with the pixels being offset by half a pixel relative to the pixels on adjacent lines. A stripe structure of R-G-B is formed on a particular line. Therefore, signal processing for such a sensor can be performed substantially in the same manner as a conventional general stripe type sensor shown in FIG. 2, excepting for the following points.
Since the sensor pixels are disposed in an offset sampling array as shown in FIG. 3, the low-pass filtering of luminance and color signals is required to be performed not one-dimensionally but two-dimensionally. This requirement will be described separately for a luminance signal and a color signal.
The switch-Y system relies upon the concept that the signal obtained from an R, G or B filter is assumed equivalently as a luminance signal. In such a system, centers (as indicated by a circle in FIG. 4) of sampling points for luminance signals Y are disposed in an offset sampling array with an offset of d/2, where d is a pitch of pixels in the horizontal direction. To obtain a high resolution by positively utilizing the advantages of the offset sampling array, it becomes necessary to properly interpolate luminance signals at positions indicated by a cross x in FIG. 4 by using adjacent pixels. An optimum interpolation value X at a position indicated by a cross x is principally an average value of all the two-dimensionally extending signals at positions indicated by a circle, the average value being weighted by an obliquely extending sine function.
There are well known simple interpolation methods of obtaining an approximate interpolation value, such as EQU X=1/4a+1/4b+1/4%c+1/4d (1) EQU X=1/2a+1/2 (2) EQU X=a (3)
From a different point of view, the above methods can be considered as performing a two-dimensional low-pass filtering by calculating a two-dimensional convolution function by inserting a sampled value into the position indicated by a circle and a zero value into the position indicated by a cross x.
The above interpolation calculation (1), (2) and (3) correspond to the following convolution functions Y1, Y2 and Y3. ##EQU1##
The following convolution function Y4, although it leads to the same result of the interpolation value X as that of the convolution function Y1, has a stronger low-pass filtering effects because the data at a position indicated by a circle is also given an average operation with respect to adjacent data. ##EQU2## In summary, forming a two-dimensional convolution function by using an apparent luminance signal obtained from an R, G or B filter for a position indicated by a circle and a zero value for a position indicated by a cross x, corresponds to performing a two-dimensional low-pass filtering for a luminance signal.
Next, a method of obtaining each color signal will be described. With the color filter array shown in FIG. 3, the array of R filters for example becomes as indicated by a circle in FIG. 5. Therefore, there arises an issue how the R color signal at a position indicated at a cross x is interpolated. To solve this problem, in the same manner as for the luminance signal, there is formed a two-dimensional convolution function by using sampled data for a position indicated by a circle and a zero value for a position indicated by a cross x. Since it is sufficient for a color signal to have a bandwidth greatly narrower than a luminance signal, it becomes necessary to low-pass filter a color signal broadly, and the number of filter stages (dimension of a convolution function) becomes large. Such a convolution function may be: ##EQU3##
The above description is also applicable to the other color signals B and G. Accordingly, a two-dimensional low-pass filtering of a color signal is performed using a convolution function such as C1 by inserting actual data for a position where data are present, and a zero value for a position where data are not present.
In order to digitally realize a two-dimensional low-pass filter of N rows and m columns, a delay for (m-1) pixels and (N-1) memories are required. For instance, if a two-dimensional low-pass filter Y4 for a luminance signal is to be realized, two 1H memories are required, as well as six 1H memories for a color signal, totaling in eight 1H memories and resulting in a problem of a considerably increased circuit scale.
The bandwidth of the low-pass filter 210 for a luminance signal is generally broader than that of the low-pass filters 205 to 207 for a color signal. In other words, if a low-pass filter is constructed of a finite response (FIR) type digital filter, the number Ny of taps of a luminance signal low-pass filter becomes smaller than the number Nc of taps of a color signal low-pass filter. Therefore, it becomes necessary to provide a delay circuit corresponding to the difference (Nc-Ny) between taps after the luminance low-pass filter to synchronize the luminance and color signals. Further, the luminance and color signal low-pass filters are generally constructed separately so that delay circuits totaling in number (Ny-1)+(Nc-1).times.3+(Nc-Ny) are required, resulting in a large circuit scale.
Consider also that there are separately formed the luminance low-pass filter 210, aperture compensation filter 212 and color low-pass filters 205 to 207. Delay circuits total in number (N1-1)+(N3-1)+3.times.(N2-1), resulting in a large circuit scale, where N1 represents the number of taps of the luminance low-pass filter, N2 represents the number of taps of each of the color low-pass filters, and N3 represents the number of taps of the aperture compensation filter.
The problems associated with the filters have been described with respect to the horizontal direction. Similar problems arise also in the vertical direction in a two-dimensional filter. In the latter case, not only the delay circuits but also the 1H memories should be provided, thus leading to a considerable increase in apparatus cost.